Research map / Image generation / GANs
Large-scale GAN latent analysis
Train high-fidelity GANs and analyze their latent semantics.
108 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in GANs, keyword cluster “synthesis · latent · face”
State History Benchmarks Tried Baselines Preprints Who Export
Where it stands
Few recent reports None of the tracked benchmarks has a report from 2025–26.
10 papers in 2023–24 and 4 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 17.2% to 16.0%.
Open questions the data points to
- Papers report AttnGAN on COCO (FID) anywhere from 35.49 to 77.01: comparisons against it depend on whose number you take.
- Papers report AttnGAN on CUB (FID) anywhere from 13.02 to 23.98: comparisons against it depend on whose number you take.
No longer reported by papers here: CIFAR-10 (FID, last in 2022).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Generative Adversarial Text to Image Synthesis (ICML 2016 · 2,214 citations)
- 2017 · Stacked Generative Adversarial Networks (CVPR 2017 · 432 citations)
- 2018 · Deformable GANs for Pose-Based Human Image Generation (CVPR 2018 · 542 citations)
- 2019 · Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · 3,479 citations)
- 2020 · Analyzing and Improving the Image Quality of StyleGAN (CVPR 2020 · 6,132 citations)
- 2021 · Alias-Free Generative Adversarial Networks (NeurIPS 2021 · 847 citations)
- 2022 · DF-GAN: A Simple and Effective Baseline for Text-to-Image Synthesis (CVPR 2022 · 314 citations)
- 2023 · Scaling Up GANs for Text-to-Image Synthesis (CVPR 2023 · 392 citations)
- 2024 · Procedural Level Generation with Diffusion Models from a Single Example (AAAI 2024 · 9 citations)
- 2025 · Exploring Sparse MoE in GANs for Text-conditioned Image Synthesis (CVPR 2025 · 2 citations)
- 2026 · Scalable GANs with Transformers (ICML 2026 · 0 citations)
Earliest papers here
- Generative Adversarial Text to Image Synthesis (ICML 2016 · 2,214 citations)
- Learning What and Where to Draw (NeurIPS 2016 · 205 citations)
- Coupled Generative Adversarial Networks (NeurIPS 2016 · 102 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
CIFAR-10 · FID (lower is better)
4 papers, 2019–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 14.73 | Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · 3,479 citations) |
| 2020 | 13.41 | Transformation GAN for Unsupervised Image Synthesis and Representation Learning (CVPR 2020 · 31 citations) |
| 2022 | 7.7 | Self-Supervised Dense Consistency Regularization for Image-to-Image Translation (CVPR 2022 · 28 citations) |
What has been tried
103 papers here report 1,105 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 6%, hurt 19%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Remove random cropping during training | Minecraft sample generation · Visual quality | Hurt | Procedural Level Generation with Diffusion Models from a Single Example (AAAI 2024 · 9 citations) | |
| Use re-captioned training data | MS COCO, 64×64 · zero-shot FID30K | 22.44 -> 16.72 | Helped | Exploring Sparse MoE in GANs for Text-conditioned Image Synthesis (CVPR 2025 · 2 citations) |
| Use code dimension 1 instead of 4 | Oxford-Dog · FID | 36.30 -> 37.93 | Hurt | Style Quantization for Data-Efficient GAN Training (CVPR 2025 · 2 citations) |
| use the 3RB+1TB architecture instead of Pix2pix's 9RB architecture | FFHQ concepts · FID | 49.14 -> 42.65 | Helped | E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · 1 citation) |
| Use hybrid decoding instead of pure sampling or greedy decoding | CVRP and TSP, varying sizes · Gap (%) | Helped | Adversarial Generative Flow Network for Solving Vehicle Routing Problems (ICLR 2025 · 0 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| AttnGAN | COCO | FID | 35.49 – 77.01 | 2 papers, e.g. CVPR 2019 and ICLR 2020 |
| AttnGAN | CUB | FID | 13.02 – 23.98 | 2 papers, e.g. ICLR 2020 and CVPR 2019 |
| StyleGAN2 | FFHQ 256×256 | FID | 3.62 – 5.14 | 2 papers, e.g. CVPR 2022 and NeurIPS 2021 |
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 14 papers; from OpenAlex and the papers' affiliations).
- Peng Cheng Laboratory · 3 papers
- Tsinghua University · 3 papers
- The Chinese University of Hong Kong · 3 papers
- Shanghai AI Laboratory · 2 papers
Use it
Download BibTeX of 26 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- GAN-based image manipulation (90 papers)
- Conditional GANs (25 papers)
- GAN vocoders and audio synthesis (13 papers)
Elsewhere in the topic
- Stabilized GAN training (144 papers)
- Generative adversarial network theory (62 papers)
- GAN latent optimization and evaluation (52 papers)
- GANs for text generation (21 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).